metalsvm/arch/x86/mm/page64.c

/* 
 * Copyright 2012 Stefan Lankes, Chair for Operating Systems,
 *                               RWTH Aachen University
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 * This file is part of MetalSVM. 
 */

#include <metalsvm/stddef.h>
#include <metalsvm/stdio.h>
#include <metalsvm/stdlib.h>
#include <metalsvm/mmu.h>
#include <metalsvm/vma.h>
#include <metalsvm/string.h>
#include <metalsvm/page.h>
#include <metalsvm/spinlock.h>
#include <metalsvm/processor.h>
#include <metalsvm/tasks.h>
#include <metalsvm/errno.h>
#include <asm/irq.h>
#include <asm/multiboot.h>
#include <asm/apic.h>

/*
 * Virtual Memory Layout of the standard configuration
 * (1 GB kernel space)
 *
 * 0x000000000000 - 0x0000000FFFFF: reserved for IO devices (16MB)
 * 0x000000100000 - 0x00000DEADFFF: Kernel (size depends on the configuration) (221MB)
 * 0x00000DEAE000 - 0x00003FFFFFFF: Kernel heap
 * 0xFF8000000000 - 0xFFFFFFFFFFFF: Paging structures are mapped in this region (max 512GB)
 */

/* 
 * Note that linker symbols are not variables, they have no memory allocated for
 * maintaining a value, rather their address is their value.
 */
extern const void kernel_start;
extern const void kernel_end;

// boot task's page map and page map lock
extern page_map_t boot_pml4;
static spinlock_t kslock = SPINLOCK_INIT;

page_map_t* get_boot_page_map(void)
{
	return &boot_pml4;
}

/** @brief Copy a single page frame
 *
 * @param src	virtual address of source page frame
 * @return	physical addr to copied page frame
 */
static size_t copy_page_frame(size_t *src)
{
	kprintf("copy_page_frame(%p)\n", src);
#if 1 // TODO: untested
	size_t phyaddr, viraddr;

	// allocate and map an empty page
	phyaddr = get_page();
	if (BUILTIN_EXPECT(!phyaddr, 0))
		return 0;

	viraddr = vma_alloc(PAGE_SIZE, VMA_HEAP);
	if (BUILTIN_EXPECT(!viraddr, 0))
		return 0;

	viraddr = map_region(viraddr, phyaddr, 1, MAP_KERNEL_SPACE);
	if (BUILTIN_EXPECT(!viraddr, 0))
		return 0;

	// copy the whole page
	strncpy((void*) viraddr, (void*) src, PAGE_SIZE);

	// unmap and free page
	unmap_region(viraddr, 1);
	vma_free(viraddr, viraddr+PAGE_SIZE);

	return phyaddr;
#else
	kprintf("TODO: copy_page_frame(%lx)\n", source);
	return 0;
#endif
}

static inline size_t canonicalize(size_t addr)
{
	if (addr & (1UL<<47))
		return addr;
	else
		return addr & ((1UL<<48) - 1);
}

static inline int map_to_level(size_t addr)
{
	if (addr >= PAGE_PML4)
		return 4;
	else if (addr >= PAGE_PDPT)
		return 3;
	else if (addr >= PAGE_PGD) 
		return 2;
	else if (addr >= PAGE_PGT)
		return 1;
	else
		return -EINVAL;
}

static inline const char * map_to_lvlname(size_t addr)
{
	const char* names[] = {"(none)", "PGT", "PGD", "PDPT", "PML4"};
	return names[map_to_level(addr)];
}

static inline size_t map_to_virt(size_t addr)
{
	return canonicalize(addr << (map_to_level(addr) * PAGE_MAP_SHIFT));
}

/*
 * Copy page maps using recursion
 *
 * @param from	pointer to virtual address of source page tables
 * @param to	pointer to virtual address of destination page tables
 * @param copy	flags what should be copied (see #define COPY_*)
 * @return	number of new allocated page frames (for tables only)
 */
static int copy_page_map(page_map_t *src, page_map_t *dest, int copy)
{
	page_map_t* next_src, * next_dest;

	int ret = 0;
	uint32_t i;
	for(i=0; i<PAGE_MAP_ENTRIES; i++) {
		if (!(src->entries[i] & PG_PRESENT))
			// skip empty entries
			dest->entries[i] = 0;
		else if (src->entries[i] & PG_USER) {
			size_t phys;
			kprintf("d:%p (%s: 0x%012lx) -> %p\n", &src->entries[i], map_to_lvlname((size_t) &src->entries[i]), map_to_virt((size_t) &src->entries[i]),  &dest->entries[i]);

			// deep copy user tables
			if ((size_t) src >= PAGE_PGT) {
				phys = get_page();
				if (BUILTIN_EXPECT(!phys, 0))
					return -ENOMEM;

				dest->entries[i] = phys|(src->entries[i] & ~PAGE_MASK);

				// reuse pointers to next lower page map tables
				next_src  = (page_map_t*) ((size_t) &src->entries[i] << 9);
				next_dest = (page_map_t*) ((size_t) &dest->entries[i] << 9);

				ret += 1 + copy_page_map(next_src, next_dest, copy);
			}
			// deep copy page frame
			else {
				if (copy) {
					phys = copy_page_frame((size_t*) src->entries[i]);
					dest->entries[i] = phys|(src->entries[i] & ~PAGE_MASK);
				}
				kprintf("c: %p (%lx)\n", &src->entries[i], src->entries[i]);
			}
		}
		// shallow copy kernel only tables
		else {
			kprintf("s:%p (%s: 0x%012lx) -> %p\n", &src->entries[i], map_to_lvlname((size_t) &src->entries[i]), map_to_virt((size_t) &src->entries[i]),  &dest->entries[i]);
			dest->entries[i] = src->entries[i];
		}
	}

	kputs("r\n");
	return ret;
}

int create_page_map(task_t* task, int copy)
{
	size_t phys;
	uint32_t ret;

	// fixed mapping for paging structures
	page_map_t *current = (page_map_t*) PAGE_PML4;
	page_map_t *new = (page_map_t*) (PAGE_PML4 - 0x1000);

	// get new pml4 table
	phys = get_page();
	if (!phys) return -ENOMEM;

	current->entries[PAGE_MAP_ENTRIES-2] = phys|KERN_TABLE;
	new->entries[PAGE_MAP_ENTRIES-1] = phys|KERN_TABLE;

	tlb_flush(); // ouch :(

	spinlock_lock(&kslock);
	ret = copy_page_map(current, new, copy);
	spinlock_unlock(&kslock);

	new->entries[PAGE_MAP_ENTRIES-1] = phys|KERN_TABLE;
	current->entries[PAGE_MAP_ENTRIES-2] = 0;

	task->page_map = (page_map_t*) phys;

	kprintf("create_page_map: allocated %u page tables\n", ret);
	return ret;
}

int drop_page_map(void)
{
#if 1
	kprintf("TODO: test drop_page_map()\n");
	return -EINVAL; // TODO
#else
	task_t* task = per_core(current_task);
	page_map_t* pml4, * pdpt, * pgd, * pgt;
	size_t phys;
	uint32_t i, j, k, l;

	pml4 = task->page_map;

	if (BUILTIN_EXPECT(pml4 == &boot_pml4, 0))
		return -EINVAL;

	spinlock_lock(&task->page_lock);

	// delete all user pages and tables
	for(i=0; i<PAGE_MAP_ENTRIES; i++) { // pml4
		if (pml4->entries[i] & PG_USER) {
			for(j=0; j<PAGE_MAP_ENTRIES; j++) { // pdpt
				if (pdpt->entries[j] & PG_USER) {
					for(k=0; k<PAGE_MAP_ENTRIES; k++) { // pgd
						if (pgd->entries[k] & PG_USER) {
							for(l=0; l<PAGE_MAP_ENTRIES; l++) { // pgt
								if (pgt->entries[l] & PG_USER)
									put_page(pgt->entries[l] & PAGE_MASK);
							}
							// TODO: put pgt
						}
					}
					// TODO: put pgd
				}
			}
			// TODO: put pdpt
		}
	}

	put_page(virt_to_phys((size_t) pml4));
	task->page_map = NULL;

	spinlock_unlock(&task->page_lock);

	return 0;
#endif
}

size_t virt_to_phys(size_t viraddr)
{
	task_t* task = per_core(current_task);
	size_t phyaddr;
	size_t* pte;

	spinlock_irqsave_lock(&task->page_lock);

	pte = (size_t *) (PAGE_PGT | (viraddr >> 9));
	phyaddr = (*pte & PAGE_MASK) | (viraddr & ~PAGE_MASK);

	spinlock_irqsave_unlock(&task->page_lock);

	return phyaddr;
}

size_t map_region(size_t viraddr, size_t phyaddr, uint32_t npages, uint32_t flags)
{
	task_t* task = per_core(current_task);
	size_t i, ret;

	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
		return 0;

	if (!viraddr) {
		kputs("map_region: deprecated vma_alloc() call from within map_region\n");
		viraddr = vma_alloc(npages*PAGE_SIZE, VMA_HEAP);
		if (BUILTIN_EXPECT(!viraddr, 0)) {
			kputs("map_region: found no valid virtual address\n");
			ret = 0;
			goto out;
		}
	}

	// correct alignment
	phyaddr &= PAGE_MASK;
	viraddr &= PAGE_MASK;
	ret = viraddr;

	if (flags & MAP_KERNEL_SPACE)
		spinlock_lock(&kslock);
	else 
		spinlock_irqsave_lock(&task->page_lock);

	kprintf("map_region: map %u pages from 0x%lx to 0x%lx with flags: 0x%x\n", npages, viraddr, phyaddr, flags);
	for(i=0; i<npages; i++, viraddr+=PAGE_SIZE, phyaddr+=PAGE_SIZE) {
		// page table entry
		size_t* pte = (size_t *) (PAGE_PGT|(viraddr >> 9));

		if (*pte && !(flags & MAP_REMAP)) {
			kprintf("map_region: 0x%lx is already mapped\n", viraddr);
			ret = 0;
			goto out;
		}

		if (flags & MAP_USER_SPACE)
			*pte = phyaddr|USER_PAGE;
		else
			*pte = phyaddr|KERN_PAGE;

		if (flags & MAP_NO_CACHE)
			*pte |= PG_PCD;

		if (flags & MAP_NO_ACCESS)
			*pte &= ~PG_PRESENT;

		if (flags & MAP_WT)
			*pte |= PG_PWT;

		if (flags & MAP_USER_SPACE)
			atomic_int32_inc(&task->user_usage);

		tlb_flush_one_page(viraddr);
	}

out:
	if (flags & MAP_KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return ret;
}

int change_page_permissions(size_t start, size_t end, uint32_t flags)
{
#if 0
	uint32_t index1, index2, newflags;
	size_t viraddr = start & PAGE_MASK;
	size_t phyaddr;
	page_map_t* pgt;
	page_map_t* pgd;
	task_t* task = per_core(current_task);

	pgd = per_core(current_task)->page_map;
	if (BUILTIN_EXPECT(!pgd, 0))
		return -EINVAL;

	spinlock_lock(&task->page_lock);

	while (viraddr < end)
	{
		index1 = viraddr >> 22;
		index2 = (viraddr >> 12) & 0x3FF;

		while ((viraddr < end) && (index2 < 1024)) {
			pgt = (page_map_t*) (page_map_t*) ((KERNEL_SPACE - 1024*PAGE_SIZE + index1*PAGE_SIZE) & PAGE_MASK);
			if (pgt && pgt->entries[index2]) {
				phyaddr = pgt->entries[index2] & PAGE_MASK;
				newflags = pgt->entries[index2] & 0xFFF;  // get old flags

				if (!(newflags & PG_SVM_INIT)) {
					if ((newflags & PG_SVM_STRONG) && !(newflags & PG_PRESENT) && (flags & (VMA_READ|VMA_WRITE) && !(flags & VMA_NOACCESS)))
						newflags |= PG_PRESENT;
					else if ((newflags & PG_SVM_STRONG) && (newflags & PG_PRESENT) && (flags & VMA_NOACCESS))
						newflags &= ~PG_PRESENT;
				}

				// update flags
				if (!(flags & VMA_WRITE)) {
					newflags &= ~PG_RW;
				} else  {
					newflags |= PG_RW;
				}

				pgt->entries[index2] = (newflags & 0xFFF) | (phyaddr & PAGE_MASK);

				tlb_flush_one_page(viraddr);
			}

			index2++;
			viraddr += PAGE_SIZE;
		}
	}

	spinlock_unlock(&task->page_lock);
#endif

	return -EINVAL;
}


int unmap_region(size_t viraddr, uint32_t npages)
{
	task_t* task = per_core(current_task);
	page_map_t* pdpt, * pgd, * pgt;
	size_t i;
	uint16_t index_pml4, index_pdpt;
	uint16_t index_pgd, index_pgt;

	if (BUILTIN_EXPECT(!task || !task->page_map, 0))
		return -EINVAL;

	if (viraddr <= KERNEL_SPACE)
		spinlock_lock(&kslock);
	else
		spinlock_irqsave_lock(&task->page_lock);

	i = 0;
	while(i<npages)
	{
		index_pml4 = (viraddr >> 39) & 0x1FF;
		index_pdpt = (viraddr >> 30) & 0x1FF;
		index_pgd = (viraddr >> 21) & 0x1FF;
		index_pgt = (viraddr >> 12) & 0x1FF;

		// currently, we allocate pages only in kernel space.
		// => physical address of the page table is identical of the virtual address
		pdpt = (page_map_t*) (task->page_map->entries[index_pml4] & PAGE_MASK);
		if (!pdpt) {
			viraddr += (size_t) PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_SIZE;
			i += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES;
			continue;
		}

		pgd = (page_map_t*) (pdpt->entries[index_pdpt] & PAGE_MASK);
		if (!pgd) {
			viraddr += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES*PAGE_SIZE;
			i += PAGE_MAP_ENTRIES*PAGE_MAP_ENTRIES;
			continue;
		}

		pgt = (page_map_t*) (pgd->entries[index_pgd] & PAGE_MASK);
		if (!pgt) {
			viraddr += PAGE_MAP_ENTRIES*PAGE_SIZE;
			i += PAGE_MAP_ENTRIES;
                        continue;
                }

		if (pgt->entries[index_pgt])
			pgt->entries[index_pgt] &= ~PG_PRESENT;

		viraddr +=PAGE_SIZE;
		i++;

		if (viraddr > KERNEL_SPACE)
			atomic_int32_dec(&task->user_usage);

		tlb_flush_one_page(viraddr);
	}

	if (viraddr <= KERNEL_SPACE)
		spinlock_unlock(&kslock);
	else
		spinlock_irqsave_unlock(&task->page_lock);

	return 0;
}

static void pagefault_handler(struct state *s)
{
	task_t* task = per_core(current_task);
	size_t viraddr = read_cr2();
	size_t phyaddr;

#if 0
	if ((viraddr >= task->start_heap) && (viraddr <= task->end_heap) && (viraddr > KERNEL_SPACE)) {
		viraddr = viraddr & PAGE_MASK;

		phyaddr = get_page();
		if (BUILTIN_EXPECT(!phyaddr, 0))
			goto oom;

		if (map_region(viraddr, phyaddr, 1, MAP_USER_SPACE) == viraddr) {
			memset((void*) viraddr, 0x00, PAGE_SIZE);
			return;
		}

		kprintf("Could not map 0x%x at 0x%x\n", phyaddr, viraddr);
		put_page(phyaddr);
	}
	/*
	 * handle missing paging structures for userspace
	 * all kernel space paging structures have been initialized in entry64.asm
	 */
	else if (viraddr >= PAGE_PGT) {
		kprintf("map_region: missing paging structure at: 0x%lx (%s)\n", viraddr, map_to_lvlname(viraddr));

		phyaddr = get_page();
		if (BUILTIN_EXPECT(!phyaddr, 0))
			goto oom;

		// TODO: initialize with zeros
		// TODO: check that we are in userspace

		// get pointer to parent page level entry
		size_t *entry = (size_t *) ((int64_t) viraddr >> 9 & ~0x07);

		// update entry
		*entry = phyaddr|USER_TABLE;

		return;
	}
#endif

	kprintf("PAGE FAULT: Task %u got page fault at %p (irq %llu, cs:rip 0x%llx:0x%llx)\n", task->id, viraddr, s->int_no, s->cs, s->rip);
	kprintf("Register state: rax = 0x%llx, rbx = 0x%llx, rcx = 0x%llx, rdx = 0x%llx, rdi = 0x%llx, rsi = 0x%llx, rbp = 0x%llx, rsp = 0x%llx\n", 
		s->rax, s->rbx, s->rcx, s->rdx, s->rdi, s->rsi, s->rbp, s->rsp);

	irq_enable();
	abort();

oom:
	kputs("map_region: out of memory\n");
	irq_enable();
	abort();
}

int arch_paging_init(void)
{
	uint32_t i, npages;

	// replace default pagefault handler
	irq_uninstall_handler(14);
	irq_install_handler(14, pagefault_handler);

	/*
	 * In longmode the kernel is already maped into the kernel space (see entry64.asm)
	 * this includes .data, .bss, .text, VGA, the multiboot & multiprocessing (APIC) structures
	 */

#if MAX_CORES > 1
	// reserve page for smp boot code
	if (!map_region(SMP_SETUP_ADDR, SMP_SETUP_ADDR, 1, MAP_KERNEL_SPACE|MAP_NO_CACHE)) {
		kputs("could not reserve page for smp boot code\n");
		return -ENOMEM;
	}
#endif

#ifdef CONFIG_MULTIBOOT
#if 0
	// map reserved memory regions into the kernel space
	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MEM_MAP)) {
		multiboot_memory_map_t* mmap = (multiboot_memory_map_t*) mb_info->mmap_addr;
		multiboot_memory_map_t* mmap_end = (void*) ((size_t) mb_info->mmap_addr + mb_info->mmap_length);

		while (mmap < mmap_end) {
			if (mmap->type != MULTIBOOT_MEMORY_AVAILABLE) {
				npages = mmap->len / PAGE_SIZE;
				if ((mmap->addr+mmap->len) % PAGE_SIZE)
					npages++;
				map_region(mmap->addr, mmap->addr, npages, MAP_KERNEL_SPACE|MAP_NO_CACHE);
			}
			mmap++;
		}
	}
#endif

	/* 
	 * Modules like the init ram disk are already loaded.
	 * Therefore, we map these modules into the kernel space.
	 */
	if (mb_info && (mb_info->flags & MULTIBOOT_INFO_MODS)) {
		multiboot_module_t* mmodule = (multiboot_module_t*) ((size_t) mb_info->mods_addr);

		npages = mb_info->mods_count * sizeof(multiboot_module_t) >> PAGE_SHIFT;
		if (mb_info->mods_count * sizeof(multiboot_module_t) & (PAGE_SIZE-1))
			npages++;
		map_region((size_t) (mb_info->mods_addr), (size_t) (mb_info->mods_addr), npages, MAP_REMAP|MAP_KERNEL_SPACE);

		for(i=0; i<mb_info->mods_count; i++, mmodule++) {
			// map physical address to the same virtual address
			npages = (mmodule->mod_end - mmodule->mod_start) >> PAGE_SHIFT;
			if (mmodule->mod_end & (PAGE_SIZE-1))
				npages++;
			kprintf("Map module %s at 0x%x (%u pages)\n", (char*) mmodule->cmdline, mmodule->mod_start, npages);
			map_region((size_t) (mmodule->mod_start), (size_t) (mmodule->mod_start), npages, MAP_REMAP|MAP_KERNEL_SPACE);
		}
	}
#endif

	// we turned on paging => now, we are able to register our task
	register_task();

	// APIC registers into the kernel address space
	map_apic();

	return 0;
}